In older machines, large pages are =2MB, and of you have enough of them, let's say 10G, you may use in my.cnf
large_pages=1
innodb_buffer_pool_size=10G
and innodb correctly allocates from this faster, never swappable memory pool.
BUT, if the machine is newer, and you booted with a kernel command line with
hugepagesz=1G default_hugepagesz=1G
Then you only need to allocate 10 pages to get 10GB of memory. This makes memory managent much faster
however, Innodb get confused. if you add
innodb_buffer_pool_size=5G
it will allocate 50G from the OS, verifiable doing
cat /proc/meminfo | grep HugePages
yet internally it will think it has only 5G, the rest is wasted.

I have a box ready to show the issue to Elena is she wants to see it. I have seen the issue in many boxes.